Abstract:A parameter design method for a full bridge LLC resonant converter is proposed, which includes magnetic inductance, resonant inductance, resonant capacitance and transformer primary to secondary side turn ratio. Through this design method, zero voltage switching (ZVS) of the primary side switching tube can be achieved effectively while obtaining a larger magnetic inductance,which means the reduced losses of the switching devices and higher converter efficiency. Moreover, the proposed method of drawing voltage gain characteristic curve clusters can intuitively and efficiently determine the inductance coefficient K and quality factor Q, optimizing parameter design. Meanwhile, theoretical analysis and experimental research are conducted on the influence of inductance coefficient K on the efficiency of the converter. Finally, the correctness and validity of this parameter design method and related theoretical research are verified by the 200 W experimental results of on a full bridge LLC prototype.

Abstract:In order to solve the problems of large voltage fluctuation and long switching time when switching between grid-connected and off-grid modes under the traditional control strategy, and the need to boost the voltage when the parallel structure is connected to the medium and high voltage power grid, a series photovoltaic DC microgrid control strategy with unified grid-connected mode was proposed. Firstly, the on-grid and off-grid structure model of the tandem photovoltaic system is established. Secondly, the DC bus voltage is regulated by droop control and PI control, and the on-grid and off-grid mode is unified through the difference between the reference power and the maximum output power, so as to realize the bus voltage stability and maximum power tracking during the switching process. Finally, the stability of the system is proved by the small signal method, and the feasibility and stability of the proposed control strategy are verified by experimental results.

Abstract:Aiming at the problem of parameter changes and coupling errors during the operation of permanent magnet synchronous motor, this paper proposes an online multi-parameter identification model based on the interconnected adaptive extended Kalman observer. First, by establishing an interconnected multi-parameter coupling compensation identification model to reduce the impact of measurement noise and parameter coupling errors on identification accuracy, high-precision identification results are obtained. Secondly, the adaptive method is used to dynamically adjust the extended Kalman observer to ensure the speed and accuracy of motor parameter identification after working conditions change, and the Lyapunov function is used to analyze the convergence of the observer when there is a model error. Finally, simulation and semi-simulation experiments were conducted on Matlab and RT-LAB semi-simulation physical system platforms. The results show that the method in this paper effectively reduces the measurement noise error and parameter coupling error, and significantly improves the anti-disturbance performance of the observer.

Abstract:In this paper, two miniaturized SIW filters in the Ku-band are investigated, and the equivalent magnetic wall cutting method is adopted to obtain the quarter mode substrate integrated waveguide in order to meet the required miniaturization specifications. Two distinct QMSIW configurations, designated as triangular QMSIW and rectangular QMSIW, are developed based on the distinct equivalent magnetic wall cuts. Subsequently, the QMSIW filters with two distinct structural configurations are subjected to investigation. However, the out-of-band rejection capability of the designed QMSIW filters is not high, and therefore the complementary split ring resonators are loaded in the QMSIW resonant cavity in order to improve the out-of-band rejection. Then we employ the electromagnetic simulation software to simulate the proposed QMSIW filters with the two different structural configurations, and obtain the ensuing results. The passband range of the rectangular QMSIW filter loading the CSRRs is from 14.94 to 16.02 GHz, with a relative bandwidth of 6.97%. The insertion loss is better than 0.6 dB, the return loss is better than 15 dB, and the out-of-band rejection is better than 35 dB@19~20 GHz, with a size of 0.2×0.314λ2g. The triangular QMSIW structure filter loading the CSRRs exhibits a passband range of 14.89~16.11 GHz, with a relative bandwidth of 7.87%. The insertion loss is better than 0.6 dB, the return loss is better than 17.2 dB, and the out-of-band rejection is better than 40 dB@19~20 GHz with a size of 0.27×0.27λ2g. Both modified QMSIW filters are processed using an alumina oxide thin film and tested with GSG probes and subsequent to the actual test, it is determined that the test results are essentially consistent with the simulation results and met the expectations.

Abstract:This paper proposes a BPPID control system based on an improved sparrow algorithm to address the problem of getting stuck in local optima when optimizing the initial weights of BPPID using the traditional sparrow algorithm. Improving population diversity by introducing composite chaotic mapping; utilizing the golden ratio and adaptive Levy flight strategy to balance the algorithm′s global search and local development capabilities; using fuzzy logic adaptive reverse learning strategy to improve the algorithm′s global search and adaptability to complex environments. The benchmark functions were tested using standard sparrow algorithm, improved sparrow algorithm, grey wolf optimization algorithm, whale optimization algorithm, improved whale optimization algorithm, particle swarm optimization algorithm, and improved particle swarm optimization algorithm to compare and verify the effectiveness of the improved sparrow algorithm. The experimental results showed that the system efficiency and fairness of the improved sparrow algorithm were superior to other algorithms. Applying the improved sparrow algorithm to solve the initial weights of BPPID in switch mode power supply systems can significantly improve the system′s dynamic response and reduce overshoot.

Abstract:In order to better solve the underdetermined anti-collision problem of RFID system, the separation algorithm is optimized from the perspective of initializing the separation matrix based on the blind source separation method. Since the mixing matrix determines the linear mapping relationship between the source signal and the observed signal, it directly affects the convergence of the separation algorithm and the quality of the separation results. Therefore, the selection of the initial mixing matrix is crucial to the performance and effectiveness of the algorithm. The initial mixing matrix is calculated using the successive nonnegative projection algorithm, which abandons the traditional random initialization and avoids the algorithm from falling into the local optimal solution. Since the tag signals of RFID are bounded, the bounded component analysis algorithm is used in the next step to separate the tag signal from the mixed signal. The simulation results show that the separation similarity of this algorithm is improved by 3.05% compared with the traditional bounded component analysis algorithm at low signal-to-noise ratio, and the separation accuracy is improved by 6.64% compared with the commonly used non-negative matrix factorization algorithm. Its low bit error rate also shows that the system can effectively handle interference and noise during data transmission or reception, thereby reducing the occurrence of data errors.

Abstract:Traditional path planning algorithms have problems such as low efficiency, easy to fall into local optimal solutions, low convergence accuracy, etc. The subtractive average optimization algorithm has fewer parameters and simpler principles than other algorithms, but it ignores the influence of optimal values during the search process, which causes the algorithm to fall into local optimal solutions. Aiming at this problem, this paper proposes a subtractive average optimization algorithm incorporating multi-strategy improvement for path planning. First of all, Tent chaotic mapping is used to initialize the search agent population to ensure the diversity of the population; an adaptive guidance mechanism is introduced to enable the algorithm to adaptively choose a better update method with the number of iterations; the population update strategy of the sine-cosine algorithm is integrated into the update method of the search agent, and the good fluctuating and oscillating nature of the sine-cosine algorithm is utilized to balance the global and local searches of the algorithm and to better ensure the algorithm′s convergence accuracy. Finally, the proposed algorithm is simulated and tested by choosing seven benchmark test functions and setting different raster map environments. The results show that the proposed algorithm has better convergence accuracy and speed, and the performance index of path planning is better and the planning effect is better.

Abstract:Distracted driving is one of the main causes of road traffic safety problems. Aiming at the problems of high computational complexity, limited generalization ability and unsatisfactory detection accuracy of existing detection algorithms based on deep learning, this paper constructs a lightweight distracted driving behavior detection algorithm based on improved YOLOv8n. Firstly, the Context Anchor Attention mechanism was fused into StarNet to form StarNet-CAA, and StarNet-CAA was integrated into the backbone network of YOLOv8n to improve the global feature extraction ability of the model and significantly reduce the computational complexity. Subsequently, FasterBlock combined with CGLU is added to the neck network to form the C2f-Faster-CGLU module, which reduces the computational cost. In addition, the shared convolution is introduced into the detection head to further reduce the computational burden and parameter size. Experimental results show that the improved YOLOv8n algorithm significantly improves the efficiency of distracted driving behavior detection, reaching an accuracy of 99.3%on the StateFarm dataset. The number of parameters of the model is reduced by 46.7%, and the amount of calculation is reduced by 41.5%. In addition, the generalization experiment is carried out on the 100-Driver dataset, and the results show that the generalization effect of the proposed scheme is improved compared with YOLOv8n. Therefore, the proposed algorithm significantly reduces the computational burden while maintaining high reliability and generalization ability.

Abstract:To meet the needs of UUV cluster collaborative communication in complex marine environments, a magnetic induction communication system based on BPSK is proposed. A channel transmission model is established to analyze the effects of different transceiver models, seawater, carrier synchronization, and carrier frequency on channel transmission, combined with MATLAB simulations and experiments for validation. Simulation results show that BPSK modulation performs well under complex channel conditions; the one-to-three receive model is insensitive to attitude changes; phase offset is a major factor in the effects of eddies, and carrier synchronization significantly reduces its impact. Ultimately, the results indicate that the bit error rate decreases to the order of 10-3, effectively meeting the collaborative communication needs of UUV clusters,validating the system′s high stability and reliability in complex marine conditions. Finally, hardware experiments further validate the feasibility of the system.

Abstract:Semantic communication is a type of communication designed to convey semantic information, which is characterized by the fact that it can effectively reduce redundancy and the amount of transmitted data. Currently the research on semantic communication is only in its infancy, and more theoretical research can help to promote the real implementation of semantic communication systems. The core technology for realizing semantic communication, end-to-end joint source channel coding, has made great progress in the past few years, and semantic images have also been developed. In order to solve the problems of computational inefficiency and insufficient semantic feature extraction, a new neural network JSCC is designed in this paper.Specifically, inspired by the excellent performance of Swin Transformer in visual tasks, the Swin-Transformer module is combined with residual networks for the first time, and a Swin Transformer-based image semantic communication system. In order to solve the problems such as the poor efficiency of traditional CNN for image feature extraction, the attention residual network module is introduced to extract the image semantic features initially, and then the image semantic features are further extracted by Swin Transformer. Through the verification of the experimental results, compared with the existing schemes, the proposed scheme in this paper achieves higher than 2 dB performance improvement in PSNR and more than 5% performance improvement in MS-SSIM performance

Abstract:With the rapid advancement of remote sensing satellite technology in China, space missions are becoming increasingly complex, posing challenges to traditional spaceborne storage and data transmission systems in terms of high customization and costly migration. This study aims to develop an integrated high-performance storage and data transmission system to address these issues. Leveraging the high flexibility of field-programmable gate arrays (FPGAs), the system design incorporates SATA III solid-state drive read/write access with a file system, multi-channel DDR controllers with data multiplexing control, and data transmission functionality supporting intermediate frequency modulation.With minimal device and software deployment, the system achieves a maximum downlink bandwidth of 900 Megabits per second and offers the most diverse data transmission capabilities. The system has been successfully implemented in a specific Jitian satellite model mission. Ground tests demonstrate that the payload storage link bandwidth remains consistently stable above 2.8 Gigabits per second, with peak bandwidth reaching 3.69 Gigabits per second. Both storage and data transmission operations achieve long-term zero-error code performance. On-orbit verification confirms clear and complete transmission of payload images. The system fully meets the stringent stability and reliability requirements of remote sensing satellites, demonstrating significant application value in the field of remote sensing satellite technology.

Abstract:There are problems such as feature imbalance and insufficient feature fusion in the visible and infrared image fusion pedestrian detection algorithm. To address the above problems, we propose a multispectral pedestrian detection network MIFNet with phased feature fusion, a dual-stream network that handles both visible and infrared inputs, an intermodal information fusion module that changes the structure of the network to reduce the impact of feature imbalance, and an extraction-injection structure that automatically learns how to extract multimodal global information during the process of feature extraction and injects it into the visible and infrared features efficiently, which improves the robustness of the network and feature fusion effect. The feature enhancement fusion module is designed and embedded to enhance the unique information of the two modalities to further improve the feature fusion effect. The experimental results show that the leakage rate of the algorithm is only 9.74%, which is 6% lower than that of the baseline algorithm, effectively improving the detection performance of the algorithm.

Abstract:In response to the problem of poor detection performance of some high-risk moving targets in autonomous driving perception tasks due to complex road environments and insufficient fusion of onboard radar and camera data, this paper designs an object detection network MLFusionNet that integrates radar and visual multi-level information based on Centerfusion. Firstly, data level fusion is added to the input layer, which concatenates the radar echo features with the image in the form of pixel values, and then inputs them into the encoding and decoding network through a secondary residual fusion module, enriching the input information of the network; then, a bottleneck structured context module was designed between the encoder and decoder of the backbone network, which obtains broader contextual information from the feature map through a multi branch convolutional structure and reduces the number of parameters through compression channels; finally, a parallel attention fusion module was designed to solve the problem of insufficient feature level modal fusion. The experimental results on the nuScenes dataset showed that the NDS of MLFusionNet reached 46.6%, which increased the mAP of cars, trucks, and pedestrians by 1.4、3.0 and 1.5 percentage points respectively compared to the multimodal network Centerfusion. This indicates that the network pays more attention to high-risk dynamic targets in the driving environment.

Abstract:A lightweight semantic segmentation model named C2LA-U2-Net, equipped with a cross attention mechanism and a residual refinement module, was proposed to address issues such as the inability to recognize fine features, blurry defect boundaries, and large model parameters in the segmentation of surface defects in polycrystalline solar cells. Firstly, a C2LA module with a cross attention mechanism was designed in the external decoding stage to extract multi-scale spatial features, reduce spatial information loss, and capture long-range dependencies, which enhanced the segmentation performance for small defects. Secondly, a lightweight twostage residual refinement module (D-RRM) was introduced to tackle the issue of blurry prediction boundaries by modeling fine-grained features to improve boundary precision. Finally, Ghost convolutions were incorporated to further reduce model complexity. Experimental results indicated that, compared to the baseline model, the C2LA-U2-Net model achieved improvements of 3.1% in mean pixel accuracy (MPA), 4.49% in mean intersection over union (MIoU), 4.39% in mean recall rate (MRecall), and 4.17% in F1 score. At the same time, the model′s parameters and GFLOPs decreased by 89.77% and 56.68%, respectively, while inference speed increased by 76.97%, demonstrating the effectiveness of the proposed method.

Abstract:Blood cell detection is a critical tool for diagnosing various diseases, as changes in blood cell count and morphology often reflect a person′s health condition. However, manual detection is time-consuming and prone to errors and omissions. To address these challenges, this paper presents an improved blood cell detection algorithm based on the YOLOv7 framework, named YOLOv7-SMC. The algorithm integrates spatial and channel reconstruction convolution to reduce feature redundancy and enhance performance. Additionally, a mixed local channel attention is incorporated in the neck network to strengthen the model′s representational capability. The algorithm also replaces the nearest neighbor interpolation upsampling with a content-aware reassembly of features upsampling operator, which adaptively adjusts the upsampling strategy to produce detailed and smooth results. Furthermore, a minimum point distance intersection over union loss function is introduced to simplify the similarity comparison between bounding boxes. Experimental results on the BCCD dataset demonstrate that this algorithm improves the mean average precision at IoU thresholds of 0.5 and 0.5:0.95 by 2.6% and 2.9%, respectively, indicating its high practicality and accuracy.

Abstract:Semantic segmentation with image-level annotation has been widely studied for its friendly annotation and satisfactory performance. Aiming at the problem of sparse activation regions and semantic ambiguity between foreground and background of class activation maps, a dual-branch weakly supervised semantic segmentation network based on activation modulation is proposed. The network uses Resnet50 and Vision Transformer as a two-branch feature extraction network, and designs an activation modulation module embedded in the convolutional branch, which forces the model to activate the intermediate fraction of pixels to generate a compact class activation map, thus alleviating the problem of sparse activation regions of class activation maps. Second, a dynamic threshold adjustment strategy based on cosine annealing decay is proposed, which adaptively determines the highest background threshold during the training process, so that more low-confidence foreground pixels are involved in the segmentation training, and complete and accurate segmentation maps are generated. The effectiveness of the network is verified on the PASCAL VOC 2012 as well as MS COCO 2014 datasets. mIou values are 74.2% and 74.0% on the PASCAL VOC 2012 validation and test sets, respectively, and 45.9% on the MS COCO 2014 validation set. The experimental results show that the network can solve the mis-segmentation problem and achieve excellent segmentation performance in scenes with similar front background colours.

Abstract:When the SLAM system estimates the camera position, a large number of feature points of moving objects participate in the feature tracking thread leading to a decrease in the accuracy and robustness of the algorithm, so how to efficiently and accurately reject the dynamic objects in the scene is particularly important. Existing dynamic vision SLAM algorithms may miss detecting or incorrectly recognize static objects as dynamic objects and reject them when dealing with dynamic objects, which triggers the problem of insufficient number of static feature points, thus affecting the stability and accuracy of the SLAM system. Therefore, this paper proposes a visual SLAM method based on panoptic segmentation and multi-view geometry, which uses panoptic segmentation FPN network to accurately recognize all objects in the segmented image, rejects a priori dynamic feature points and retains as many static features as possible, based on which LK optical flow method with fused image pyramid is used to realize optical flow tracking and reject parallel dynamic feature points, and potential dynamic feature points are used to track the dynamic feature points. The potential dynamic feature points are rejected more effectively by the multi-view geometry method based on dynamic probability, which avoids the omission of dynamic feature points and realizes the comprehensive screening of dynamic objects in the scene to improve the accuracy of the system. The construction of semantic map and octree map is realized on the basis of sparse point cloud constructed by the system. The experiments use the TUM RGB-D dataset to verify the system localization accuracy, and the results show that the root mean square error (RMSE) of the absolute trajectory error of this algorithm is reduced by an average of 84.34% in all sequences compared with ORB-SLAM2, which significantly improves the robustness and accuracy of the system,and it is of use to construct two maps that can be used for SLAM upper layer tasks.

Abstract:To address the challenges of multiscale features and complex background processing in plant pest and disease detection in modern agriculture, this paper proposes an efficient and accurate detection model, AgriSwin, to improve the precision and efficiency of agricultural pest and disease detection. The AgriSwin model is based on the Swin Transformer and integrates a dilated feature aggregation module and an adaptive spatial convolution module. The dilated feature aggregation module extracts multi-scale features through convolutional layers with different dilation rates and optimizes feature fusion using an adaptive weighting mechanism for global feature information. The adaptive spatial convolution module generates adaptive weights to dynamically weight the feature maps, enhancing the ability to capture both local and global information in complex backgrounds. Experimental results show that the AgriSwin model achieves detection accuracies of 79.65%、99.90%、and 95.08% on the PlantDoc, PlantVillage, and custom datasets, respectively. Additionally, the model′s parameter count is reduced by 25.63% compared to Swin Transformer-T, significantly lowering memory and computational resource requirements while maintaining high accuracy, demonstrating its broad potential for large-scale agricultural applications.

Abstract:To solve the problems of complex frequency information, strong time variation and obvious modulation characteristics of planetary gearbox vibration signal, a fault diagnosis method of planetary gearbox based on visual spectral feature fusion was proposed. Initially, Welch′s transformation is applied to planetary gearbox signals to obtain power spectra. Subsequently, a visual graph algorithm is used to construct a graph spectrum, and centrality measures of the graph nodes are calculated to form a feature matrix. Finally, an improved CNN-Inception model is employed to obtain the fault diagnosis results of the planetary gearbox. Experimental results demonstrate that this method can accurately identify faults in planetary gearboxes. In the experimental datasets covering two operational conditions, the model achieves an accuracy of 98.57%, demonstrating its generalization ability. Compared with alternative methods, the proposed approach exhibits higher accuracy and stronger generalization capabilities.

Abstract:The maintenance and prediction of turbofan engine lifespan are critical to modern aviation, playing a key role in ensuring safety and minimizing operational costs. This study addresses the challenge of predicting the RUL of turbofan engines by proposing a novel hybrid model that integrates Parallel TCN and Bidirectional BiLSTM. Traditional methods often struggle to capture both local features and long-term dependencies simultaneously; the proposed model overcomes this limitation by using TCN to extract short-term local features and BiLSTM to capture bidirectional temporal dependencies. To further improve feature importance recognition, an enhanced SE attention mechanism is introduced, which dynamically adjusts feature weights to better highlight critical information. Experiments conducted on the FD001 and FD003 subsets of the C-MAPSS dataset demonstrated that the proposed model achieved RMSE values of 12.15 and 11.16, and Scores of 230.4 and 209.84, respectively, outperforming other approaches in terms of accuracy.

Abstract:In order to solve the lack of accuracy of data feature extraction caused by missing values, nonlinear and non-stationary characteristics in the measurement data of oil well dynamic liquid level, and the problem that the accurate measurement of oil well dynamic liquid level position cannot be achieved, an abnormal identification method of oil well dynamic liquid level measurement data based on normalized RBF neural network is proposed. Through the sensor installed on the oil well to collect data in real time, the multi-source oil normalization processing technology based on expert database is used to complete the data verification and integration. Empirical mode decomposition (EMD) is used to decompose the data into trend and fluctuation terms. After removing the fluctuation terms, the trend data is used as the input of normalized RBF neural network. The experimental results show that this method can effectively complete incomplete data, accurately identify abnormal data through the trend term and provide reasonable alternative values, and the obtained dynamic liquid level position curve is basically consistent with the actual dynamic liquid level position curve, with the maximum error of less than 2 m, which can realize the accurate estimation of the dynamic liquid level position of oil wells. The abnormal identification method of oil well dynamic liquid level measurement data based on normalized RBF neural network solves the challenges brought by data missing, nonlinearity and non stationarity, realizes the accurate estimation of oil well dynamic liquid level position, and provides technical support for real-time monitoring and data analysis of oil well dynamic liquid level.